CN103810499B - Application for detecting and tracking infrared weak object under complicated background - Google Patents
Application for detecting and tracking infrared weak object under complicated background Download PDFInfo
- Publication number
- CN103810499B CN103810499B CN201410062826.0A CN201410062826A CN103810499B CN 103810499 B CN103810499 B CN 103810499B CN 201410062826 A CN201410062826 A CN 201410062826A CN 103810499 B CN103810499 B CN 103810499B
- Authority
- CN
- China
- Prior art keywords
- entropy
- target
- image
- tracking
- infrared
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
- Aiming, Guidance, Guns With A Light Source, Armor, Camouflage, And Targets (AREA)
Abstract
The invention discloses an application for detecting and tracking an infrared weak object under a complicated background. The application is characterized by comprising the following steps of: 1, suppressing clutters and keeping the topological structure of an image, and constructing a bionic vision weighted entropy model with an adjacent airspace and a preferred direction to realize conversion for the image from a grey mode to an entropy model; 2, analysing the movement state of the weak object with burst and stationary characteristics, and constructing a self-adaptive entropy flow target movement estimation model meeting the maneuvering features of the weak object by virtue of the nonlinear diffusion smoothing and self-adaptive local restriction criterion of an entropy flow to realize the approximation of an estimation speed to the real movement state of the weak object; 3, searching a weak object tracking method adopting generic multi-feature and measurement, and constructing a multi-feature fused sequential filter model to realize accurate, robust and real-time identification for the weak object. The invention discloses a self-adaptive entropy flow detection and tracking algorithm for the infrared weak object, and enriches a detection and tracking technology for the weak object.
Description
Technical field
The present invention relates to Infrared images pre-processing technology, the analysis of weak signal target kinestate and multiple features fusion and degree
The weak signal target tracking field of amount, and in particular to infrared Faint target detection and the application for tracking under complex background.
Background technology
The background technology of the present invention is related to three committed steps and method:The Preprocessing Technique of holding topological structure,
The method for tracking target of the target state, similar multiple features fusion and tolerance of analysis burst or smooth performance.
Keep the Infrared images pre-processing technology of topological structure:
The presence of clutter and noise, they improve the difficulty of Faint target detection.Background is suppressed using spatial filtering method
With noise, the topological structure of its image can produce change.The suppressing method of research clutter and noise, the letter for both having improved weak signal target are miscellaneous
More again than the topological structure for keeping image.
Analysis burst or the target state of smooth performance:
The motion of weak signal target has burst or smooth performance, for characterizing uncertainty of objective and mobility.Foundation
Apparent caused by image entropy pattern is exactly to portray target motion, for area pixel gray scale is gentle or the characteristics of acute variation,
The smooth target motion estimation model with constraint criterion of research, realization are approached to target state.
Similar multiple features fusion and the method for tracking target measured:
The presence of pseudo- target, and target occlusion, intersection, separation, appearance, the phenomenon for disappearing, they improve target following
Difficulty.The target typically generic character with identical or convergence, studies the target following side of generic multiple features fusion and tolerance
Method can be rejected pseudo- target and realize the tracking of weak signal target.
The content of the invention
It is an object of the invention to provide infrared Faint target detection and the application of tracking under complex background, keeps infrared weak mesh
The preconditioning technique of mark sequence image topological structure, clutter reduction and raising target signal to noise ratio;Build according to entropy model and meet weak
The estimation model of the self adaptation entropy flow target motion of target maneuver feature, approaches the kinestate of target;And it is many using generic
Feature Fusion and the method recognition and tracking weak signal target measured, it is achieved thereby that infrared Faint target detection and tracking.
By setting up weighting entropy model, the target motion estimation model of self adaptation entropy flow and the Sequential filter of multiple features fusion
The goal in research of model, identification weak signal target and weak signal target track.The present invention needs the key scientific problems for solving as follows:
(1) build the spatial domain mask of bionical thing vision significance tolerance.Mask projects the neighbouring spy with orientation preferentially in spatial domain
Levy, using the Weighted information entropy matrix description gradation of image information matrix of spatial domain mask, realize the conversion of image entropy pattern.
(2) build the adaptive variation model of Nonlinear Diffusion.Smoothness is controlled using the Nonlinear Diffusion factor, is adapted to
The data item and the scale factor of smooth item of adjustment entropy flow Variation Model, realizes that speed to be estimated approaches the kinestate of target.
(3) build target association function.In set spatial domain and time domain, the seriality of target motion and concordance, with
And the homogeny of target generic, determine that it will be occurred in adjacent domain with maximum probability.Correlation function comprising spatial domain and time domain,
Motion and non-athletic generic character, realize target detection using multiple features Distance evaluation function.
The technical solution adopted for the present invention to solve the technical problems is:
The method of the present invention includes following key step:
1st, suppress noise, clutter and keep the topological structure of image, build the neighbouring bionical vision with orientation preferentially in spatial domain
Weighting entropy model, realizes that image is transformed to entropy pattern from grayscale mode.
2nd, apparent motion by caused by image entropy pattern is exactly to portray weak signal target motion, analysis burst or smooth performance
Target state.According to the nonlinear smoothing and local restriction criterion of entropy flow, using the smooth journey of Nonlinear Diffusion factor control
Spend, and entropy flow is constrained and smooths set of constraints contract beam speed to be estimated, obtain the entropy flow field of densification, structure meets target maneuver
The self adaptation entropy flow target motion estimation model of feature, convergence target state.3rd, generic multiple features fusion is studied with tolerance
Method for tracking target, build the Sequential filter model of multiple features fusion, realize that weak signal target is accurate, robust with track in real time.
Its infrared Faint target detection is as shown in Figure 1 with trace flow figure.
It is an advantage of the invention that:
(1) propose the neighbouring weighting entropy model with orientation preferentially in spatial domain
Background and noise are suppressed using space filtering generally, it can easily lose the information of target and change the topology knot of image
Structure.Suppress the problem of background and noise for the present invention, watch mechanism attentively according to biology, construct spatial domain it is neighbouring with orientation preferentially plus
Power spatial domain mask.The comentropy tolerance image pixel gray level weighted using spatial domain mask, realizes that image is transformed to by grayscale mode
Entropy pattern.It provides a kind of new approaches with the topological structure for keeping image for clutter reduction.
(2) propose the target motion estimation model of the Nonlinear Diffusion of self adaptation entropy flow
The uncertainty of weak signal target and the interference of mobility and clutter to estimation, they can reduce target motion and estimate
The degree of accuracy of meter.Carry out the research contents of estimating target motion state for the present invention, it is gentle or drastically in the face of Entropy Changesization
Feature, self-adaptative adjustment entropy model data item and the scale factor of smooth item, smooth journey using the control of the Nonlinear Diffusion factor
Degree, realizes entropy flow constraint and smooths set of constraints contract beam speed to be estimated, with convergence target state.It is raising weak signal target
The degree of accuracy of estimation provides a kind of new method.
Description of the drawings
Fig. 1 is the infrared Faint target detection and trace flow figure of the present invention;
Fig. 2 is the Faint target detection and tracking technique scheme of the present invention;
Fig. 3 is technology path of the present invention by grayscale mode to entropy mode conversion;
Fig. 4 is the weak signal target motion estimation techniques route in the present invention.
Specific embodiment
The present invention is using infrared Faint target detection under complex background as shown in Figure 2 and tracking technique scheme, its concrete reality
Apply step as follows:
(1) visual fixations embody base sketch map the Strength Changes acutely position at place and its geometric distribution and organizational structure, and
The discontinuity point of primitive normal direction and each point in the depth, depth of observer, the discontinuity point on surface normal direction.
Neuron receptive field adopts Gauss-exponential model, and ganglionic cell biography is described by overlap with one heart, deformation area of different sizes
The center of system receptive field, perimeter region, area of disinthibiting on a large scale, wherein Gauss model state center and perimeter region successively, refer to
Exponential model describes area of disinthibiting on a large scale.Gauss-index mathematical model referring to formula (1), by two Gauss models and one
Exponential model is superimposed, and obtains space filtering mask, so as to solve the spatial domain mask of the bionical vision significance tolerance of key issue
Design, target spatial domain is neighbouring to select feature with orientation preferentially to submit to it, realizes the purpose of clutter reduction and noise jamming.
Wherein A1、A2、A3The peak factor at central authorities, surrounding and edge, σ are represented respectively1、σ2Represent respectively central, surrounding
Scale coefficient, and fy、fxVertical direction, the gradient of horizontal direction are represented according to this.
When the model is used for processing luminance contrast edge, edge contrast can be strengthened well, quilt can be effectively lifted again
Regional luminance contrast and half tone information that heritage is filtered by the Yezhong heart.
(2) in theory of information, image is considered the carrier of information aggregate.Shannon entropy has description information amount or uncertainty
Characteristic, the quantity of information of image subblock is described using the weighting Shannon entropy of spatial domain mask.From left to right, use successively from top to bottom
Weighted information entropy measures image pixel gray level, the entropy diagram complete so as to build a width.It will solve the conversion of image model, and which adds
Power entropy mathematical model is referring to formula (2):
Wherein f (x, y) represents gray scale of the image at (x, y), and ρ (x, y) represents probability density function
, G (x, y) representation space filter mask, J × K represent image local window size, and M × N represents picture size.In meter
When calculating comentropy, in ρ (x, y) definition, factor 1/e is multiplied by, to ensure Shannon entropy ρ (x, y) log ρ (x, y) interval [0,1/
E] monotonic increase.
Using technology path (1) and (2), the neighbouring weighting entropy model with orientation preferentially in spatial domain is built, realizes image from ash
Degree mode conversion is entropy pattern.Its technology path is as shown in Figure 3.
(3) the strong direction of Entropy Changesization should not add smoothness constraint, and smoothness constraint should be added in perpendicular to the side of gradient
Upwards.Data constraint item is only set up in entropy diagram gradient larger part simultaneously.Use data constraint at the larger point of gradient, and
Smoothing Constraint is used only at the less point of gradient.Weight function is defined, when gradient is more than a certain threshold value, weight function is 1, using data
Constraints;When less than a certain threshold value, weight function is 0, data constraint condition is closed;Other situations, using partial weighting data
Constraints.It is by the ratio of the self-adaptative adjustment data item and smooth item captured in key scientific problems adaptive variation model
The design of the factor, its mathematical model is referring to formula (3):
Wherein P represents entropy diagram picture, and (u, v) represents entropy flow field, and ψ (x, y) represents weight function,Represent the ladder of entropy diagram picture
Degree modular function, C (u, v) represent data item, and S (u, v) represents smooth item.
(4) for self adaptation entropy flow motion estimation model smooth item, in the relatively flat region in entropy flow field, conduct because
Son can be automatically increased so that the less random fluctuation in flat region is smoothed, and near the mutation of entropy flow field, transduction factors energy
Automatically it is less, then edge is then barely affected.Transduction factors as the nonlinear function for successively decreasing, using Nonlinear Diffusion because
Son control smoothness, its mathematical model is referring to formula (4):
Wherein V=(u, v) represents entropy flow field, and R (u, v) represents diffusion depending on the data item in formula (3), D (u, v)
, depending on the combination that item and the Nonlinear Diffusion factor are smoothed in formula (3).It will solve the adaptive variation of Nonlinear Diffusion
The structure of the Nonlinear Diffusion factor in model.Using technology path (3) and (4), the Nonlinear Diffusion of self adaptation entropy flow is built
Target motion estimation model.By the optimization solution of steepest descent method, realize that entropy flow approaches weak signal target state.Its technology path is such as
Shown in Fig. 4.
(5) candidate target and its kinestate are identified from the target motion estimation model of self adaptation entropy flow, using fortune
It is dynamic to carry out Expressive Features collection with non-athletic information.Meeting the adjacent domains premise of spatial feature, by synthesis strategy will move with it is non-
Motion feature merges, and builds the Multiple feature association Distance evaluation function of target.It may be then target to meet appreciation condition, then is passed through
The energy accumulation Sequential filter method of multiple image target trajectory is recognizing weak signal target and target following.
Claims (4)
1. the detection of infrared weak signal target and tracking under complex background, is characterized in that method and step is as follows:
(1) Preprocessing Technique of topological structure, the Gaussian function being distributed using different scale is kept to be superimposed with exponential function
Gaussian index mixed model is generated, to construct the sky with the neighbouring bionical vision significance tolerance with local orientation feature in spatial domain
Domain mask, according to any pixel in mask window and the Euclidean distance of center pixel, distributes the coefficient weights of the mixed model, its
Shown in Gaussian index mixed model such as formula (1):
Wherein:A1、A2、A3The peak factor of representative function, σ1、σ2The scale coefficient of Gaussian function is represented, x and y is represented respectively and covered
The horizontal and vertical distance of any pixel and center pixel, f in mould windowxAnd fyRepresent respectively horizontally and vertically
Gradient;
Then the quantity of information of image subblock is described using spatial domain mask Weighted information entropy, first, any pixel is calculated and is gone out in image
Existing gray probability;Secondly, the local message entropy that Shannon entropy calculates the image subblock is weighted using spatial domain mask;Finally, calculate
The local message entropy of all image subblocks simultaneously builds entropy diagram picture in order;So as to realize by gray level image mode conversion be entropy diagram as mould
Formula, is allowed to the topological structure for not only improving the signal to noise ratio of weak signal target but also keeping image;The mathematical table of its gray probability, weighting Shannon entropy
Up to formula such as formula (2) Suo Shi:
Wherein:M × N and J × K represent infrared image size and infrared image sub-block window size, f (x, y) and ρ (x, y) respectively
The gray probability that image occurs in the gray scale of (x, y) and at (x, y) is represented respectively, and G (x, y) represents spatial domain mask weight coefficient, H
Represent weighting Shannon entropy;To ensure Shannon entropy function limit, formula is multiplied by 1/e in (2), to avoid changing entropy picture structure
Information;
(2) burst or the easy motion state of weak signal target are analyzed, according to entropy flow data and smooth constraint criterion, using formula (3)
Build a class entropy flow Variational Functional;According to the difference of each pixel gradient of entropy diagram picture, the weight function value of adaptive polo placement pixel
ψ (x, y), when gradient is more than a certain threshold value, weight function value is more than 0 and is less than or equal to 1, using data constraint condition, conversely, power
Functional value is 0, closes data constraint condition;And according to the difference of each pixel gradient of entropy diagram picture, using gradient modular function L (| | ▽ P
| |) adaptively control entropy flow smoothness;Formula (3) solves the extreme value of Variational Functional using steepest descent method, to approach
The movement velocity of weak signal target and the kinestate for describing the target;Wherein:P represents entropy diagram picture, and (u, v) represents entropy flow, C (u, v)
Data item is represented, S (u, v) represents smooth item;
(3) the weak signal target tracking of similar multiple features fusion and tolerance, the target typically generic with identical or convergence are studied
Feature, using motion and the non-athletic feature set of weak signal target, analyzes multiple features synthesis strategy, and the distance for building Multiple feature association is commented
Valency function, identification weak signal target and target trajectory.
2. a kind of infrared Faint target detection according to claim 1 and tracking, it is characterised in that:It is described to keep topology
The Infrared images pre-processing technology of structure, builds the neighbouring bionical visual weight entropy model with orientation preferentially in spatial domain, changing image
Pattern, realizes suppressing noise, clutter and keeps the purpose of image topology structure.
3. a kind of infrared Faint target detection according to claim 1 and tracking, it is characterised in that:Described image entropy mould
Apparent motion caused by formula come portray target motion, analysis burst or smooth performance weak signal target kinestate;According to entropy flow
Nonlinear smoothing and local restriction criterion, smoothness, and entropy flow constraint peace are controlled using the Nonlinear Diffusion factor
The estimating speed of the self adaptation combination constraint weak signal target of sliding constraint, obtains the entropy flow field of densification, realizes estimating speed approaching to reality
Weak signal target kinestate.
4. a kind of infrared Faint target detection according to claim 1 and tracking, it is characterised in that:The motion feature
Tracking be difficult to be adapted to the infrared track applied environment of complexity, using the similar target typically class with identical or convergence
Category feature, builds the method for tracking target of generic multiple features fusion of the motion with non-athletic and tolerance, rejects pseudo- target, recognize weak
Target and target trajectory.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410062826.0A CN103810499B (en) | 2014-02-25 | 2014-02-25 | Application for detecting and tracking infrared weak object under complicated background |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410062826.0A CN103810499B (en) | 2014-02-25 | 2014-02-25 | Application for detecting and tracking infrared weak object under complicated background |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103810499A CN103810499A (en) | 2014-05-21 |
CN103810499B true CN103810499B (en) | 2017-04-12 |
Family
ID=50707240
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410062826.0A Expired - Fee Related CN103810499B (en) | 2014-02-25 | 2014-02-25 | Application for detecting and tracking infrared weak object under complicated background |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103810499B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111259942A (en) * | 2020-01-10 | 2020-06-09 | 西北工业大学 | Method for detecting weak target in water |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104268844B (en) * | 2014-10-17 | 2017-01-25 | 中国科学院武汉物理与数学研究所 | Small target infrared image processing method based on weighing local image entropy |
CN104992452B (en) * | 2015-06-25 | 2019-02-22 | 中国计量学院 | Airbound target automatic tracking method based on thermal imaging video |
CN107657626B (en) * | 2016-07-25 | 2021-06-01 | 浙江宇视科技有限公司 | Method and device for detecting moving target |
CN107280673B (en) * | 2017-06-02 | 2019-11-15 | 南京理工大学 | A kind of infrared imaging breath signal detection method based on key-frame extraction technique |
CN108364277A (en) * | 2017-12-20 | 2018-08-03 | 南昌航空大学 | A kind of infrared small target detection method of two-hand infrared image fusion |
CN110706231B (en) * | 2019-11-12 | 2022-04-12 | 安徽师范大学 | Image entropy-based three-dimensional culture human myocardial cell pulsation characteristic detection method |
CN111311191B (en) * | 2020-02-25 | 2023-05-05 | 武汉轻工大学 | Method and device for obtaining raw material quality range based on wine product quality range |
CN115249254B (en) * | 2022-09-21 | 2022-12-30 | 江西财经大学 | Target tracking method and system based on AR technology |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102289681A (en) * | 2011-08-05 | 2011-12-21 | 上海邮政科学研究院 | Method for matching envelope images |
CN102722890A (en) * | 2012-06-07 | 2012-10-10 | 内蒙古科技大学 | Non-rigid heart image grading and registering method based on optical flow field model |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030202701A1 (en) * | 2002-03-29 | 2003-10-30 | Jonathon Schuler | Method and apparatus for tie-point registration of disparate imaging sensors by matching optical flow |
-
2014
- 2014-02-25 CN CN201410062826.0A patent/CN103810499B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102289681A (en) * | 2011-08-05 | 2011-12-21 | 上海邮政科学研究院 | Method for matching envelope images |
CN102722890A (en) * | 2012-06-07 | 2012-10-10 | 内蒙古科技大学 | Non-rigid heart image grading and registering method based on optical flow field model |
Non-Patent Citations (1)
Title |
---|
《联合熵流与边缘算子的红外运动目标检测》;王忠华等;《华中科技大学学报(自然科学版)》;20110131;第39卷(第1期);论文第1-4页 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111259942A (en) * | 2020-01-10 | 2020-06-09 | 西北工业大学 | Method for detecting weak target in water |
CN111259942B (en) * | 2020-01-10 | 2022-04-26 | 西北工业大学 | Method for detecting weak target in water |
Also Published As
Publication number | Publication date |
---|---|
CN103810499A (en) | 2014-05-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103810499B (en) | Application for detecting and tracking infrared weak object under complicated background | |
CN106408594B (en) | Video multi-target tracking based on more Bernoulli Jacob's Eigen Covariances | |
CN107423698B (en) | A kind of gesture estimation method based on convolutional neural networks in parallel | |
CN104915970B (en) | A kind of multi-object tracking method based on Track association | |
CN104318258B (en) | Time domain fuzzy and kalman filter-based lane detection method | |
CN105608456B (en) | A kind of multi-direction Method for text detection based on full convolutional network | |
CN106920243A (en) | The ceramic material part method for sequence image segmentation of improved full convolutional neural networks | |
CN107967451A (en) | A kind of method for carrying out crowd's counting to static image using multiple dimensioned multitask convolutional neural networks | |
CN103729854B (en) | A kind of method for detecting infrared puniness target based on tensor model | |
CN107481188A (en) | A kind of image super-resolution reconstructing method | |
CN110119687A (en) | Detection method based on the road surface slight crack defect that image procossing and convolutional neural networks combine | |
CN105787439A (en) | Depth image human body joint positioning method based on convolution nerve network | |
CN107742099A (en) | A kind of crowd density estimation based on full convolutional network, the method for demographics | |
CN103996049B (en) | Ship overlength and overwidth detection method based on video image | |
CN107689052A (en) | Visual target tracking method based on multi-model fusion and structuring depth characteristic | |
CN104408747B (en) | Human motion detection method suitable for depth image | |
CN105957356B (en) | A kind of traffic control system and method based on pedestrian's quantity | |
CN110276785A (en) | One kind is anti-to block infrared object tracking method | |
CN106204484A (en) | A kind of traffic target tracking based on light stream and local invariant feature | |
CN109002848A (en) | A kind of detection method of small target based on Feature Mapping neural network | |
CN111709300A (en) | Crowd counting method based on video image | |
CN103871062A (en) | Lunar surface rock detection method based on super-pixel description | |
CN110070565A (en) | A kind of ship trajectory predictions method based on image superposition | |
CN105741276A (en) | Ship waterline extraction method | |
CN102289822A (en) | Method for tracking moving target collaboratively by multiple cameras |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170412 Termination date: 20200225 |
|
CF01 | Termination of patent right due to non-payment of annual fee |